Paper Title
The Design and Implementation of a GaN-Based Tuner for Adjusting Parameters of a Controlling Algorithm

Abstract
In Industry Evolution 4.0, smart factories adopt modern controlling algorithms on sophisticated machines for better production quality and yield. Experts intend to leverage novel artificial intelligence techniques to design such controlling algorithms. However, when applying these algorithms to sophisticated machines, a factory manager is still needed to tune the controlling algorithms with their expertise in these algorithms and experience in operating the machines. The high learning curve of related knowledge and experience has jeopardized and slowed down the modern controlling algorithm deployments to sophisticated machines in the current stage. To mitigate this issue, this research proposes GANBaTe, an automatic tuner for the controlling algorithms based on the Generative Adversarial Networks (GAN). GANBaTe consists of two neural networks, NN-Tuner and NN- Simulator. NN-Tuner finds proper parameters for a controlling algorithm, and NN-Simulator simulates the performance of the sophisticated machine running the controlling algorithm. Without expertise and experience, GANBaTe can learn the machine's behavior patterns and find the most appropriate parameters for the algorithm. Through the interactive training of the two networks, GANBaTe only needs a small amount of pre-collected data to optimize the algorithm. We conduct several experiments to verify the effectiveness and characteristics of GANBaTe. We discuss the impact of various settings of GANBaTe on the tuning parameters and the resulting performances. The results show that GANBaTe reduces Sum Absolute Error (SAE) from 150 to 27 and eliminates Percentage Overshoot (PO) to less than 25%. We also prove that GANBaTe can be deployed to different controlling algorithms and machines. Keywords - Controller Parameters, Controller Tuning, Artificial Neural Networks, Generative Adversarial Networks.